2

i ran exploratory factor analysis and according to eigenvalue > 1, 41 factors were extracted out of 142 items. but when i ran MAP and parallel analysis, 16 factors were prescribed as a proper number of factors to retain. now what should I do? With 16 factors, my total variance is 51% and with 41 factor, total variance is 75% but with too many trivial items/factors. Which method should I adhere to? Eigenvalue or MAP and parallel analysis?

My second question is that i read about another method named comparison data (Ruscio and Roche, 2012) which outperform other methods, but i cant find any information about how it works, which software should I use or any manual about it?

Gala
  • 8,323
  • 2
  • 28
  • 42
Alireza aminaee
  • 53
  • 7
  • 1
    Of possible interest: [VSS criterion for the number of factors (in R's psych package)](http://stats.stackexchange.com/q/32669/930). – chl Sep 12 '13 at 11:32
  • i just want to know when one method show me 16 factors and the other one tells me 41 factors, can i choose from 16 to 41 on my wish in the way that both total variance and item loaded in each factor become proper. (for example can i choose 25 factors? or 30? something between these two? or i must adhere exactly to one of these methods? – Alireza aminaee Sep 12 '13 at 19:07
  • 1
    I'm curious as to the kind of questionnaire you are using: 41 factors for a total of 142 items means you may have factors with very few items, as you pointed out. I would neither trust the % of explained variance, nor Kaiser's rule (which is known to overestimate the number of factors). Parallel analysis should be ok, providing your factors make sense. What kind of factors extraction are you using (ML, principal components, principal axis, etc.)? Please note that Ruscio and Roche's code is available as [R code](http://www.tcnj.edu/~ruscio/taxometrics.html#EFAwithCD). – chl Sep 12 '13 at 21:19
  • you are right @chi. most of my factors have few items. i try most of the method just for comparison such as ML, PCA, PAF, all of them extracted 41 factors. except that in PCA with 75% variance but in ML and PAf with 65% variance. i know that when the research purpose is to find latent variable, one must use Factor analysis. but i want to know, is PCA used for Construct validity, too? or it is just a method of items reduction? i dont know if my question make sense. – Alireza aminaee Sep 13 '13 at 06:23
  • 1
    One thing to realize is that your work is obviously very exploratory. You should probably with what makes sense and not “adhere” blindly to any one criterion. – Gala Sep 13 '13 at 15:31
  • please, give more insight – Alireza aminaee Sep 13 '13 at 15:57

0 Answers0